Model Following Sliding Mode Control
tive of the system state and does not use time delayed signals. In
Based on Uncertainty and
Disturbance Estimator
S. E. Talole
e-mail: setalole@hotmail.com
S. B. Phadke
e-mail: sbphadke@hotmail.com
Defence Institute of Advanced Technology,
Girinagar, Pune,
Maharashtra 411025, India
A new design of sliding mode control based on an uncertainty and
disturbance estimator (UDE) is given. The control proposed does
not require the knowledge of bounds of uncertainties and distur-
bances and is continuous. Thus, two main difficulties in the design
of sliding mode control are overcome. Furthermore, the method of
UDE is extended to plants having significant uncertainty in the
control input matrix and subjected to disturbances that nonlin-
early depend on states. DOI: 10.1115/1.2909604
Keywords: sliding mode control (SMC), uncertainty and distur-
bance estimator (UDE), uncertain systems
1 Introduction
Sliding mode control SMC is an effective strategy for control-
ling systems with significant uncertainties and unmeasurable dis-
turbances. In the conventional SMC systems, insensitivity to un-
certainties and disturbances is guaranteed by employing a
discontinuous control based on the bounds of uncertainties and
disturbances. In many situations, these bounds are hard to find,
which result in an overestimation and consequently a large con-
trol. The discontinuous control is objectionable because it can
cause excessive wear and tear of actuators and may excite unmod-
eled dynamics.
One method to overcome the need to know the bounds of the
uncertainty explicitly is to find the bounds adaptively and then use
them in the SMC. Slotine and Coetsee 1 proposed an SMC, in
which the bounds on uncertain parameters are found adaptively.
Simple control laws, in which a function bounding the uncertainty
is found adaptively, are given in Refs. 2,3.
Youcef Toumi and Ito 4 proposed a nonadaptive approach
called the time delay control TDC for uncertain systems. In this
approach, the uncertainties and disturbances are estimated from
past measurements under the assumption that the uncertainties and
disturbances do not significantly change in a small interval of
time. The estimate is then used in control so as to nullify the effect
of uncertainty and disturbances on the system. Copious applica-
tions of this method including those to SMC are reported in the
literature in the past 15 years 510. In a similar approach, Chan
11 designed a control compensating the perturbation in the sys-
tem. A common concern in all these methods is the need for the
derivative of the state or the sliding surface variable.
Recently, Zhong and Rees 12 have proposed a very promising
method called the uncertainty and disturbance estimator UDE
for control of linear time invariant systems. This method, although
Contributed by the Dynamic Systems, Measurement, and Control Division of
ASME for publication in the JOURNAL OF DYNAMIC SYSTEMS, MEASUREMENT, AND
CONTROL. Manuscript received January 11, 2005; final manuscript received July 11,
2007; published online April 24, 2008. Assoc. Editor: Prabhakar R. Pagilla.
Journal of Dynamic Systems, Measurement, and Control
Copyright 2008 by ASME
Downloaded 25 Apr 2008 to 121.241.65.34. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
based on an idea similar to the TDC, does not require the deriva-
this paper, the results of Ref. 12 are extended to SMC. The
proposed control has several merits over both the conventional
SMC and the UDE method.
In conventional SMC, the control is discontinuous, which re-
sults in undesirable chatter. Furthermore, the control can be de-
signed if the bounds of uncertainty and disturbances are known.
The chattering control is undesirable for several reasons and the
bounds of uncertainty and disturbances are not always easy to
find. The proposed control enforces sliding mode without using
discontinuous control and without requiring the knowledge of un-
certainties or their bounds.
Zhong and Rees 12 have proposed a control for linear sys-
tems, which have no uncertainties in the input matrix. In this
paper, the method is extended to disturbances containing state
dependent nonlinearities and to systems having significant uncer-
tainties in the control input matrix. Furthermore, the choice of the
sliding surface proposed in this paper avoids the difficulty of large
initial amplitudes of the input seen with the UDE approach of Ref.
12. This paper analyzes the accuracy of estimation and proposes
a method to improve the accuracy. This paper is organized as
follows: Section 2 states the problem. A model following control
is designed in Sec. 3 and is illustrated by a numerical example in
Sec. 4. The conclusions are stated in Sec. 5.
2 Statement of the Problem
Consider a single input single output plant
x = Ax + bu + Ax + bu + dx,t 1
where x is the state vector, u is the control input, A and b are
known constant matrices, A and b are uncertainties, and dx , t
is an unknown disturbance.
ASSUMPTION 1. The uncertainties A and b and the distur-
bance dx , t satisfy the matching conditions given by
A = bD, b = bE, dx,t = bvx,t 2
where D and E are unknown matrices of appropriate dimensions
and vx , t is an unknown function. The system of Eq. (1) can now
be written as
x = Ax + bu + bex,t 3
where ex , t = Dx + Eu + vx , t. Although ex , t contains uncer-
tainty and disturbances, in the sequel, it will be referred to as the
lumped uncertainty for the sake of convenience. Next, let
xm = Amxm + bmum 4
be a stable model satisfying certain structural conditions stated by
the following assumption:
ASSUMPTION 2. The choice of the model is such that
A Am = bL, bm = bM 5
where L and M are suitable known matrices.
The objective is to design a control u so as to force the plant of
Eq. 3 to follow the model of Eq. 4 in spite of the uncertainties
and disturbances represented by ex , t. Equations 2 and 5 are
well-known matching conditions required to guarantee invariance
and are an explicit statement of the structural constraint stated in
Ref. 12.
3 Design of Control
In this section, a model following control is designed in the
framework of SMC. The method is based on a sliding surface
suggested in Ref. 13 and is different from the SMC based model
following reported in literature.
Define a sliding surface,
MAY 2008, Vol. 130 / 034501-1
 + e(x, t) u = ueq + un 11
1
bT b (s + k)Gf (s) Selecting
-
k
ueq = Lx + Mum 12
b Tb

un where k is a positive constant, from Eqs. 11, 12, and 10, we

Gf (s) get
= bTbun + bTbex,t k 13
Fig. 1 Block diagram representation of Eq. 15 Next, the component un will be designed.
3.1 Compensation of Uncertainties and Disturbances. The
lumped uncertainty ex , t can be compensated by estimating it on
= b Tx + z 6 the lines of Ref. 12. Rewriting Eq. 13 as

where 1
ex,t = + k un 14
b Tb
z = bTAmx bTbmum, z0 = bTx0 7
it can be seen that the lumped uncertainty ex , t can be computed
Equation 7 for the auxiliary variable z given here is different
from that given in Ref. 13. By virtue of the choice of the initial from the right hand side of Eq. 14. This, however, cannot be
condition on z, = 0 at t = 0. If a control u can be designed ensur- done directly. Let the estimate of the lumped uncertainty, denoted
ing sliding, then = 0 implies by ex , t, be defined as

x = Amx + bmum 8 1
ex,t = s + kG f s G f sun 15
b Tb
and thus fulfills the objective of model following. Differentiating
Eq. 6 and using Eq. 3 with Eq. 7 give where G f s is a strictly proper low-pass filter with unity steady-
state gain and broad enough bandwidth. It can be seen that Eq.
= b Ax + b bu + b bex,t b Amx b bmum
T T T T T
9
15 uses variables in both time and s-domain. While such a use is
not uncommon in the literature, it may be stated that the equation
=bTbLx bTbMum + bTbu + bTbex,t 10
can be interpreted as signals operating on hardware described by
Let the required control be expressed as transfer functions, as illustrated in Fig. 1. Such an interpretation

3
x 10 15
6

5 10
4
5
3

2 0
control

5
0

1 10

2
15
3

4 20
0 2 4 6 8 10 12 0 2 4 6 8 10 12
(a) time in sec (b) time in sec

2 6

1.5 5

1
4
0.5
xm

disturbance d

3
and

0
2
0.5
x

1
1

1.5 0

2 1
0 2 4 6 8 10 12 0 2 4 6 8 10 12
(c) time in sec (d) time in sec

Fig. 2 Case 1plots of the model following: a , b control input, c states of

plant and model, d disturbance

034501-2 / Vol. 130, MAY 2008 Transactions of the ASME

Downloaded 25 Apr 2008 to 121.241.65.34. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 0.025 60

0.02
40
0.015

0.01 20

0.005
0

control
0

20
0.005

0.01 40
0.015
60
0.02

0.025 80
0 2 4 6 8 10 12 0 2 4 6 8 10 12
(a) time in sec (b) time in sec

2 6

1.5 5

1
4
0.5
xm

disturbance d
3
and

0
2
0.5
x

1
1

1.5 0

2 1
0 2 4 6 8 10 12 0 2 4 6 8 10 12
(c) time in sec (d) time in sec

Fig. 3 Case 2increased uncertainty b = 0 0.8T: a , bcontrol input, c

states of plant and model, d disturbance

affords simplification of equations on the lines of block diagram 3.3 Accuracy of Estimation. The above result is based on the
simplification. With such a filter, premise that Eq. 16 holds. In this section, we take a closer look
at the error in estimation.
ex,t ex,t 16 We consider a practical low-pass filter
which enable the design of un as
1
un = ex,t 17 G f s = 23
Ts + 1
1 where T is a small positive constant. With the above G f s and in
= s + kG f s + G f sun 18
b Tb view of Eqs. 14, 15, and 20,
Solving for un gives

un = T
1
s + kG f s 19
ex,t = 1 G f s 1
b Tb
+ k un 24
b b1 G f s
Clearly, since G f s is strictly proper, the control signal un in Eq. =Tex,tG f s 25
19 is implementable.
Therefore, Eq. 16 will hold, if the term Tex , t is sufficiently
3.2 Existence of Sliding Mode. The existence of the sliding small. Interestingly, this is similar to the usual assumption
mode can be proved easily. We define the error in estimation as ex , t ex , t L, where L is a small interval of time, found in the
TDC approach.
ex,t = ex,t ex,t 20
It is worth noting that for G f s in Eq. 23, the control un
Using Eq. 17 in Eq. 13, we get works out to
= k + bTbe
which, in view of Eq. 16, leads to
x,t 21
un = T
1
b bT
k

+
s
26

= k2 22 and has a simple time domain interpretation. From Eqs. 26 and

Since = 0 at t = 0 by virtue of the choice of in Eqs. 6 and 7, 25, it is clear that smaller T implies a smaller estimation error
satisfaction of Eq. 22 ensures = 0 for all t 0. This makes the but a larger magnitude of control if sigma is not small. The choice
uncertain plant follow the stable model chosen by the designer for of as given in Eq. 6 and 7 enables the designer to strike a
all t 0. favorable compromise in this respect.

Journal of Dynamic Systems, Measurement, and Control MAY 2008, Vol. 130 / 034501-3

Downloaded 25 Apr 2008 to 121.241.65.34. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 3
x 10 25
8
20
6
15

4 10

control
2
0

0 5

10
2
15
4
20

6 25
0 2 4 6 8 10 12 0 2 4 6 8 10 12
(a) time in sec (b) time in sec

2 6

1.5 5

1 4

0.5 3
xm

disturbance d
and

0 2

0.5 1
x

1 0

1.5 1

2 2
0 2 4 6 8 10 12 0 2 4 6 8 10 12
(c) time in sec (d) time in sec

Fig. 4 Case 3a more rapidly changing disturbance: a , b control input, c

states of plant and model, d disturbance

3.3.1 Improvement of Accuracy. The accuracy of estimation

can be improved as much as desired by an appropriate choice of
G f s. For example, if Tex , t is not sufficiently small, it can be
b =
0
0.4
, vx,t = 2sintx21 + costx2 + 1

accounted for by the following filter: The model to be followed is

G f s =
1
T2s2 + 2Ts + 1
27 Am = 0
4 2.8
1
, bm =
0
4
The lumped uncertainties and disturbances can be written as There is an initial condition mismatch between the plant and the
ex,t = ex,tG f s + ex,t1 G f s model given by x0 = 1 1T and xm0 = 0 0T. The low-pass fil-
ter is chosen as 1 / 1 + Ts with T = 0.001 and the control gain k
= ex,tG f s + ex,t2Ts + T2s2G f s = 5. The reference input um is a square wave switching at 6 s. The
= ex,t1 + 2TsG f s + T2G f sex,t simulation results in Fig. 2 show that in spite of significant uncer-
tainties and disturbances, the system follows the model very
Now, defining closely. In the illustration, these results are referred to as Case 1.
ex,t = ex,t1 + 2TsG f s Next, the uncertainty in the control input matrix was increased

to 80% by taking b = 0 0.8T, while the uncertainty A was
1 kept the same as in Case 1. The simulation results shown in Fig. 3
= 1 + 2TsG f s + k un
b Tb show that the control effort has increased but the model following

accuracy is almost the same as that of the case with b = 0
s+k
= 1 + 2TsG f s un 28 0.4T. This is Case 2.
b Tb Finally, the effectiveness of the controller was assessed for a
the control un based on the above ex , t is implementable since disturbance that changed significantly faster than in Cases 1 and 2.
1 + 2Tss + kG f s is proper for the choice of G f s in Eq. 27. The disturbance term for this case is given by vx , t
Now, Eq. 16 will hold if T2ex , t is sufficiently small. The result = 2sin10tx21 + cos10tx2 + 1. The results shown in Fig. 4 con-
can be easily generalized. firm the following good model. Without going into a detailed
comparison, it appears that the method of UDE copes with fast
4 Illustrative Example varying disturbances and large control input matrix uncertainties
better than the TDC 14,15. This is referred to as Case 3. Simu-
The plant is of the form of Eqs. 1 and 2 with

lation with pure SMC was also conducted for the plant described
0 1 0 0 0 above. The results not shown here confirmed the drawbacks of
A= , b= , A = , pure SMC, as mentioned in Sec. 1.
2 3 1 2 3

034501-4 / Vol. 130, MAY 2008 Transactions of the ASME

Downloaded 25 Apr 2008 to 121.241.65.34. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 4
x 10 ates significant uncertainties, including those in control input ma-
12 trix, and disturbances. This paper also suggests how the accuracy
of the estimator can be improved.
10
References
1 Slotine, J. J. E., and Coetsee, J. A., 1986, Adaptive Sliding Controller Syn-
8 thesis for Non-Linear Systems, Int. J. Control, 436, pp. 16311651.
2 Yoo, D. S., and Chung, M. J., 1992, A Variable Structure Control law With
Simple Adaptation Laws for Upper Bounds on the Norm of Uncertainties,
6 IEEE Trans. Autom. Control, 376, pp. 159165.
3 Wheeler, G., Su, C. Y., and Stepanenko, Y., 1998, A Sliding Mode Controller

4 certainties, Automatica, 3412, pp. 16571661.
2 133142.
0 6 Morioka, H., Wada, K., and abanovic, A., 1996, Sliding Mode Control
2
0 2 4 6 8 10 12
time in sec
Fig. 5 Improvement of accuracy with second order filter
As indicated in Sec. 3.3.1, the accuracy of estimation can be
improved by employing a filter like that in Eq. 27. The reduction
of with such a modification for the plant with the same rapidly
changing disturbance as in Case 3 is shown in Fig. 5. The reduc-
tion is significant.
5 Conclusion
In this paper, uncertainty and disturbance estimator UDE is
extended and applied to SMC of uncertain plants overcoming two
main problems in the design of SMC. The effectiveness of the
proposed controller is verified by simulation. The controller toler-
With Improved Adaptation Laws for the Upper Bounds on the Norm of Un-
4 Youcef-Toumi, K., and Ito, O., 1990, A Time Delay Controller for Systems
With Unknown Dynamics, ASME J. Dyn. Syst., Meas., Control, 1121, pp.
5 Elmali, H., and Olgac, N., 1992, Sliding Mode Control With Perturbation
Estimation SMCPE: A New Approach, Int. J. Control, 564, pp. 923941.
Based on the Time Delay Estimation, IEEE Workshop on Variable Structure
Systems, pp. 102107.
7 Youcef-Toumi, K., and Shortlidge, C., 1991, Control of Robotic Manipulators
Using Time Delay, IEEE International Conference on Robotics and Automa-
tion, pp. 23912398.
8 Park, J. Y., and Chang, P. H., 2004, Vibration Control of a Telescopic Handler
Using Time Delay Control and Commandless Input Shaping Technique, Con-
trol Eng. Pract., 12, pp. 769780.
9 Kim, N. I., Lee, C. W., and Chang, P. H., 1998, Sliding Mode Control With
Perturbation Estimation: Application to Motion Control to Parallel Manipula-
tor, Control Eng. Pract., 6, pp. 13211330.
10 Xu, J., and Cao, W., 2000, Synthesized Sliding Mode and Time-Delay Con-
trol for a Class of Uncertain Systems, Automatica, 3612, pp. 19091914.
11 Chan, S. P., 1996, An Approach to Perturbation Comensation for Variable
Structure Systems, Automatica, 323, pp. 469473.
12 Zhong, Q.-C., and Rees, D., 2004, Control of Uncertain LTI Systems Based
on an Uncertainty and Disturbance Estimator, ASME J. Dyn. Syst., Meas.,
Control, 1264, pp. 905910.
13 Ackermann, J., and Utkin, V. I., 1998, Sliding Mode Control Design Based on
Ackermanns Formula, IEEE Trans. Autom. Control, 432, pp. 234237.
14 Corradini, M. L., and Orlando, G., 1998, Variable Structure Control of Dis-
cretized Continuous Systems, IEEE Trans. Autom. Control, 439, pp. 1329
1334.
15 Paranjpe, N. V., and Phadke, S. B., 2005, Comments on Variable Structure
Control of Discretized Continuous-Time Systems, IEEE Trans. Autom. Con-
trol, 504, pp. 538540.

Journal of Dynamic Systems, Measurement, and Control MAY 2008, Vol. 130 / 034501-5

Downloaded 25 Apr 2008 to 121.241.65.34. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm